HPC Efficiency: Profiling, EAR dashboard
Who is this for?
This session will present the main features of the Energy Aware Runtime (EAR) software, specifically targeted for users of high performance computing data-centers that use both HPC and AI applications. EAR is the energy management software, which is installed in several European Data Centers such as SuperMUC-NG (LRZ), Snellius (SURF) and now in MN5 (BSC).
What will you learn?
EAR offers services for system energy monitoring (including system and node powercap), job-level energy and performance monitoring, and energy optimization through a runtime library. Examples of common HPC/AI applications will be given in order to show users how to do job-level energy accounting, energy and performance monitoring, application characterization, and finally energy optimization of their application. Specifically we will include scientific applications that represent common CPU intensive, memory intensive, and GPU use cases.
Why is this important?
We will highlight how a user can monitor the energy usage of these applications using EAR with the inclusion of simple Slurm environment variables. During the tutorial we will show users how to correlate application metrics with energy results, helping users to understand their applications. Some examples of EAR data visualization will be done using Graphana and EAR job analytics tools.
How is the session structured?
Mixture of presenter led explanations and hands-on exercises. Users will have access to the National Supercomputer Snellius, therefore gaining experience with a state-of-the-art computing system. Already installed applications will be used but we will consider running user installed applications if there is enough time. Some exercises will be done by attendees but some others (such as manually changing the CPU frequency) are restricted to potential cluster limitations.
Prerequisite
Attending (or knowledge of the content) in: